When AI Adoption Misfires: What We Can Learn from Klarna’s Rollout

The deeper challenge of adoption isn’t technical. It’s behavioral.

At a Glance: When Klarna replaced 700 customer service agents with AI, it wasn’t just a cost-cutting move, it was a high-stakes bet on efficiency in the face of mounting pressure. But the rollout backfired, degrading customer experience, eroding employee morale, and forcing the company to quietly rehire humans a year later.

This article dissects what went wrong, not with the technology, but with the way it was framed and adopted. Drawing from behavioral economics and first-hand observation inside one of the country’s most respected financial firms, it explores why AI succeeds only when it supports, rather than replaces, the human systems it enters.

The lesson for leaders is clear: successful AI adoption isn’t just a technical decision. It’s a cultural one. And getting it right means understanding how people react to change, where trust lives inside organizations, and why strategic speed often starts with thoughtful design.

In moments of financial pressure, the appeal of AI becomes almost irresistible. Faster service, lower cost, and round-the-clock consistency. AI promises exactly what every CEO is tasked with delivering: more revenue, lower expense, fewer risks.

Ostensibly.

Let me tell you the story of Klarna, the Swedish fintech that helped launch the “buy now, pay later” boom during the COVID-19 pandemic. As e-commerce surged and consumers sought more flexible payment options, Klarna thrived. Revenues jumped by as much as 40%, lifting its valuation to over $45 billion.

But by 2023, shopping behaviors normalized and Klarna’s margins tightened. Especially in Europe, where interest rates were higher and macroeconomic stress more acute, financial pressure mounted. Credit losses were rising. Klarna’s pandemic-era growth story was fading.

Once Europe’s most valuable fintech, its valuation collapsed by nearly 70%. Meanwhile, competitors like Affirm and PayPal were gaining ground with more advanced underwriting, broader integrations, and stronger strategic positioning. Klarna needed to act — and quickly. In that moment, AI was more than an efficiency play. It was a life preserver.

So in 2023, leadership replaced 700 customer service agents with a generative AI assistant. The system could handle inquiries in 35 languages. But when it couldn’t resolve an issue, customers weren’t smoothly handed to a human. Instead, they had to repeat themselves — frustrated, unheard, and now angrier than before.

The result wasn’t just a degraded experience. It may have deepened reliance on human agents rather than reduced it. For the agents who remained, the job got more taxing. Layoffs had already strained morale. Now, they bore the emotional weight of conversations AI had inflamed.

In April 2025, Klarna quietly delayed its U.S. IPO. A month later, the CEO publicly acknowledged the misstep. The company began rehiring human agents to restore a support model that would now prioritize human interaction.

It’s tempting to frame AI adoption as a technical upgrade. It’s really a behavioral one.

When change moves fast, it often outruns the people it's meant to support. Under pressure, decision-makers tend to focus on what’s visible — cost curves, staffing ratios, projected efficiencies — while overlooking the slower, harder-to-measure effects on trust, resilience, and morale.

At Klarna, that pressure made the decision to scale AI feel not just rational, but obvious. Replace people. Expand coverage. Cut cost.

But strategy rarely succeeds in the abstract. It succeeds (or breaks!) in the lived experience. Customers don’t notice operational savings. They notice whether their issue gets resolved. Employees don’t feel margin improvements. They feel whether they’re supported or left to absorb the fallout of a decision they didn’t make.

The real question isn’t whether AI can create value. It’s whether the organization is behaviorally ready to adopt it in ways that protect and extend the human systems it depends on.

Behavioral economics helps explain what Klarna and many others miss about strategic adoption

There’s no question the case for AI is compelling. When technologies with sweeping potential arrive, leaders often turn to the rational case: forecasts, efficiencies, the promise of scale.

But adoption never happens in a vacuum. It unfolds inside individuals, shaped by perception, emotion, and informal influence. That’s where behavioral economics becomes essential. It helps surface what formal logic often misses.

In Klarna’s case, three behavioral dynamics may have clouded their adoption, obscuring critical signals that were easy to miss.

1. Loss aversion

Behavioral economists Daniel Kahneman and Amos Tversky found that people experience the pain of losses more intensely than the satisfaction of gains. But in practice, it’s not always about the numbers. It’s about what those losses represent.

For Klarna, the collapse in valuation and rising credit losses weren’t just economic indicators. They were a threat to the company’s identity it had built during the pandemic, when revenue surged and it emerged as Europe’s top fintech. That success became something to defend. The urgency wasn’t just to cut costs. It was to stop the slide, to prove that leadership was still in control.

Replacing 700 agents with AI made that urgency visible. It saved money, sent a signal, and promised scalability. But what didn’t show up on the balance sheet was the erosion of trust, continuity, and human care that had once made Klarna’s service model stand out.

2. Action bias

In moments of urgency, speed can feel like strength. Behavioral researchers call it action bias, the tendency to respond to uncertainty with decisive movement even before the full picture is clear. In high-accountability environments, especially at the executive level, the pressure to demonstrate control often overrides the instinct to pause, assess, or adapt.

That’s the pattern Klarna may have fallen into. With valuation in freefall and competitors gaining ground, leadership couldn’t afford to appear idle. The launch of its AI assistant was sweeping and fast. On the surface, it signaled efficiency and innovation. Internally, though, the signs of cultural resistance were likely already forming but easy to miss in the rush to execute.

When customers reached the limits of what the AI could handle, their frustration compounded. Human agents were stepping into emotionally charged moments with no context and little support. The pressure that AI was meant to relieve didn’t disappear. It landed on the people left to clean up after the system faltered.

3. The empathy gap

Research in behavioral science, especially the work of George Loewenstein and Deborah Small, suggests that people often misjudge how much emotional nuance is needed in service interactions. From a distance, these moments look procedural. But up close, they’re personal.

AI’s appeal lies in its consistency, its ability to follow rules without drift. But consistency doesn’t always translate into care. People don’t just want answers. They want to feel heard, especially in moments of financial stress or uncertainty.

Klarna’s AI held the promise of flawless execution. But it couldn’t empathize. And that difference, easy to miss in planning meetings or rollout metrics, became painfully clear in practice.

To Klarna’s credit, they adjusted. Just maybe too late.

In May 2025, the company began rehiring human agents quietly shifting toward a more flexible support model. It wasn’t a rejection of AI. It was a reframing.

Too often, organizations approach AI as a replacement strategy. But technology rarely thrives on its own. It depends on the human systems around it: the trust loops, emotional texture, and social cues that shape how people interpret an interaction.

The risks of poor integration — from relying on AI beyond its limits, breaking context, or eroding trust — aren’t signs that your AI strategy is failing. They’re signs that adoption was framed narrowly, without the behavioral lens needed to make it sustainable.

Behavioral economics doesn’t fix the technology. It helps design the environment. And when customer experience, brand trust, and employee morale are at stake, that design quietly determines whether adoption works or quietly unravels.

There is another path. I’ve seen it up close.

A few years ago, I had a front-row seat inside one of the country’s top-performing financial institutions as it rolled out similar technologies in its agent call centers. This organization was consistently recognized for service quality and customer trust. Which may have contributed to its cautious roll out and, ultimately, to a more durable application.

AI was designed to support the background actions that took agents away from the conversation. During live calls, the system would listen in and surface helpful prompts in real time, such as a disclosure reminder here, a flagged transaction there, a gentle nudge to follow up on a potential issue. After the call, it offered structured feedback, delivered directly to the agent, not filtered through a manager.

The technology was sophisticated. But what made it effective was what we call ‘fit for purpose’. There were no mass layoffs. As people left through natural attrition, fewer were hired back. Morale stayed steady. So did trust. Customer satisfaction held. Reputation never wavered.

It wasn’t a growth story. It was a stability story. AI helped reduce cost and risk without trading away the human connection.

From a behavioral perspective, that’s what successful adoption looks like: change that feels additive, not extractive. When AI is used as scaffolding rather than substitution it gives people the clarity, context, and confidence to do their best work.

That’s where Klarna’s approach broke down. By positioning AI to stand in for people, it shifted the nature of the interaction altogether. The result wasn’t just a service gap. It was a breakdown in the social architecture that made the service work in the first place.

AI can be transformative. But only when it’s applied with care to the human systems it enters. And it appears Klarna is now on that path.

The challenge with AI isn’t its power. It’s our posture toward it.

Throughout history, each leap in capability — from the printing press or the calculator — has been met with unease. A fear that something human might be lost. And in some ways, that fear isn’t wrong. Each tool reshaped how we work, how we think, and what we value. But over time, they also made space for new kinds of creativity, analysis, and connection.

Many CEOs now see AI as essential to long-term competitiveness. But most acknowledge they aren’t fully ready. According to PwC’s latest CEO survey, four in ten believe their business won’t survive another decade without transformation. But it’s not technology that will determine who makes it. It’s readiness.

AI will change the nature of work. But it doesn’t have to diminish the people doing it.

The real question isn’t just where AI can go technically. It’s how we choose to apply it and what behaviors, experiences, and relationships we want to protect along the way.

Strategic adoption isn’t about implementing tools. It’s about designing transitions. Behavioral economics offers a way to see those transitions more clearly and to understand the friction, the trust, and the quiet social signals that shape how change actually happens.

When we get that right, AI can strengthen the systems that still depend on people. And that’s the opportunity. Not to replace what makes us human, but to build around it. Carefully.

References

Blackman, R., & Sipes, T. (2024, March–April). Organizations aren’t ready for the risks of agentic AI. Harvard Business Review. https://hbr.org/2025/06/organizations-arent-ready-for-the-risks-of-agentic-ai

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291. https://doi.org/10.2307/1914185

Kim, BJ., Kim, MJ. & Lee, J. The dark side of artificial intelligence adoption: linking artificial intelligence adoption to employee depression via psychological safety and ethical leadership. Humanit Soc Sci Commun 12, 704 (2025). https://doi.org/10.1057/s41599-025-05040-2

Klarna. (2023, August 16). Klarna AI assistant handles two-thirds of customer service chats in its first month. Klarna Newsroom. https://www.klarna.com/international/press/klarna-ai-assistant-handles-two-thirds-of-customer-service-chats-in-its-first-month

Loewenstein, G., & Small, D. A. (2007). The scarecrow and the tin man: The vicissitudes of human sympathy and caring. Review of General Psychology, 11(2), 112–126. https://doi.org/10.1037/1089-2680.11.2.112

PricewaterhouseCoopers. (2025). 28th Annual Global CEO Survey. PwC. https://www.pwc.com/gx/en/issues/c-suite-insights/ceo-survey.html

Rifkind, H. (2025, June 8). The smarter AI gets, the more stuff it makes up. The Times. https://www.thetimes.com/comment/columnists/article/the-smarter-ai-gets-the-more-stuff-it-makes-up-zcjv5k6lv

Rifkind, H. (2025, June 11). Using ChatGPT for work? It might make you more stupid. The Times. https://www.thetimes.com/uk/technology-uk/article/using-chatgpt-for-work-it-might-make-you-more-stupid-dtvntprtk

Simons, C. (2024, June 5). Klarna IPO S-1 breakdown. Mostly Metrics. https://www.mostlymetrics.com/p/klarna-ipo-s1-breakdown

Twig. (2024). What Klarna got wrong about AI in customer support—and how they fixed it. https://www.twig.so/blog/what-klarna-got-wrong-about-ai-in-customer-support--and-how-they-fixed-it

Klarna’s story is a compelling case of what can happen when we miss the early signals of cultural friction. But that’s a deeper story of its own. If you’re interested in the psychology behind resistance and how to spot it before it slows your next initiative, you might find this piece insightful: Why We Resist Change (And How to Overcome It)